keras入门(二) VGG网络实现猫狗大战

2022-02-15 13:19 256阅读 0赞

上次用keras实现了简单的线性方程，接下来实现比较经典的CNN网络-----VGG16，下面显示的是VGG网络的结构图

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3UwMTI2MDE2NDc_size_16_color_FFFFFF_t_70][]

这里使用vgg-16,经过多个（卷积层，池化层），最后通过三个全连接层变为一个一维的数据，用softmax生成每个标签的类别，直接上代码，数据直接可以下载kaggle的猫狗数据集

from keras.models import Sequential
    from keras.layers import Conv2D, MaxPooling2D, Dense, Dropout, Activation, Flatten, BatchNormalization
    from keras.optimizers import SGD, Adam
    import os
    import argparse
    import random
    import numpy as np
    from scipy.misc import imread, imresize
    from keras.utils import to_categorical
    from keras.datasets import mnist
    
    parser = argparse.ArgumentParser()
    parser.add_argument('--train_dir', default='./kaggle/train/')
    parser.add_argument('--test_dir', default='./kaggle/test/')
    parser.add_argument('--log_dir', default='./')
    parser.add_argument('--batch_size', default=8)
    parser.add_argument('--gpu', type=int, default=-1)
    args = parser.parse_args()
    
    os.environ['CUDA_VISIBLE_DEVICES'] = str(args.gpu)
    type_list = ['cat', 'dog']
    
    def creat_vgg_16_net():
        model = Sequential()
        model.add(Conv2D(64, (3, 3), input_shape=(224, 224, 1), padding='same', activation='relu', name='conv1_block'))
        model.add(Conv2D(64, (3, 3), activation='relu', padding='same',name='conv2_block'))
        model.add(BatchNormalization())
        model.add(MaxPooling2D(pool_size=(2, 2), strides=(2, 2)))
        model.add(Conv2D(128, (3, 3), activation='relu', padding='same',name='conv3_block'))
        model.add(Conv2D(128, (3, 3), activation='relu', padding='same',name='conv4_block'))
        model.add(BatchNormalization())
        model.add(MaxPooling2D(pool_size=(2, 2), strides=(2, 2)))
        model.add(Conv2D(256, (3, 3), activation='relu', padding='same',name='conv5_block'))
        model.add(Conv2D(256, (3, 3), activation='relu', padding='same',name='conv6_block'))
        model.add(Conv2D(256, (1, 1), activation='relu', padding='same',name='conv7_block'))
        model.add(BatchNormalization())
        model.add(MaxPooling2D(pool_size=(2, 2), strides=(2, 2)))
        model.add(Conv2D(512, (3, 3), activation='relu', padding='same',name='conv8_block'))
        #model.add(Dropout(0.25))
        model.add(Conv2D(512, (3, 3), activation='relu', padding='same',name='conv9_block'))
        model.add(Conv2D(512, (1, 1), activation='relu', padding='same',name='conv10_block'))
        model.add(BatchNormalization())
        model.add(MaxPooling2D(pool_size=(2, 2), strides=(2, 2)))
        model.add(Conv2D(512, (3, 3), activation='relu', padding='same',name='conv11_block'))
        model.add(Conv2D(512, (3, 3), activation='relu', padding='same',name='conv12_block'))
        model.add(Conv2D(512, (1, 1), activation='relu', padding='same',name='conv13_block'))
        model.add(BatchNormalization())
        model.add(MaxPooling2D(pool_size=(2, 2), strides=(2,2)))
        model.add(Flatten())
        model.add(Dense(2048, activation='relu'))
        #model.add(BatchNormalization())
        model.add(Dropout(0.5))
        model.add(Dense(4096, activation='relu'))
        #model.add(BatchNormalization())
        model.add(Dropout(0.5))
        model.add(Dense(1, activation='sigmoid'))
        #model.add(Dense(2, activation='softmax'))
        return model
    
    
    def prepare_data():
        file_list = os.listdir(args.train_dir)
        random.shuffle(file_list)
        train_num = int(len(file_list) * 0.8)
        validation_num = len(file_list) - train_num
        train_file_list = file_list[0:train_num]
        validation_file_list = file_list[train_num:]
        return create_generate(train_file_list, args.batch_size, (224, 224)), create_generate(validation_file_list, args.batch_size, (224, 224))
    
    
    def create_generate(train_file_list, batch_size, input_size):
        while(True):
            random.shuffle(train_file_list)
            image_data = np.zeros((batch_size, input_size[0], input_size[1], 1), dtype='float32')
            label_data = np.zeros((batch_size, 1), dtype='int32')
            for index, file_name in enumerate(train_file_list):
                image = imresize(imread(args.train_dir + file_name, mode='L'), input_size)
                label = file_name.split('.')[0]
                image_data[index % batch_size] = np.reshape(image / 255, (input_size[0], input_size[1], 1))
                label_data[index % batch_size] = type_list.index(label)
                if(0 == (index + 1) % batch_size):
                    #label_data = to_categorical(labei_data, 2)
                    yield image_data, label_data
                    #label_data = keras.utils.to_categorical(label_data, 2)
                    image_data = np.zeros((batch_size, input_size[0], input_size[1], 1), dtype='float32')
                    label_data = np.zeros((batch_size, 1), dtype='int32')
    
    
    def train(model):
        sgd = SGD(lr=0.001, decay=1e-8, momentum=0.9, nesterov=True)
        #sgd = Adam(lr=0.001)
        #model.compile(optimizer=sgd, loss='categorical_crossentropy', metrics=['accuracy'])
        model.compile(optimizer=sgd, loss='binary_crossentropy', metrics=['accuracy'])
        train_generate, validation_genarate = prepare_data()
        model.fit_generator(generator=train_generate, steps_per_epoch=1200, epochs=50, verbose=1,
                            validation_data=validation_genarate, validation_steps=50, max_queue_size=1,
                            shuffle=True)
        model.save_weights(args.log_dir + 'model.h5')
    
    
    def predict(model):
        file_list = os.listdir(args.test_dir)
        cat_real_count = 0
        dog_real_count = 0
        cat_predict_count = 0
        dog_predict_count = 0
        for index, file_name in enumerate(file_list):
            label = file_name.split('.')[0]
            if 'cat' == label:
                cat_real_count = cat_real_count + 1
            else:
                dog_real_count = dog_real_count + 1
            image = imresize(imread(args.test_dir + file_name, mode='L'), (224, 224))
            label = model.predict(np.reshape(image/255, (1, 224, 224, 1)))
            print(str(label))
            if label <= 0.5:
                cat_predict_count = cat_predict_count + 1
            else:
                dog_predict_count = dog_predict_count + 1
        print(cat_real_count, cat_predict_count, cat_predict_count/ cat_real_count, dog_real_count, dog_predict_count, dog_predict_count / dog_real_count)
    
    
    if __name__ == '__main__':
        try:
            model = creat_vgg_16_net()
            #train(model)
            model.load_weights('E:/private/deeplearning/model.h5')
            predict(model)
        except Exception  as err:
            print(err)

训练过程中，验证集的准确率一直维持在75%，但是测试的时候准确率却更高，不知道什么原因，有大神的话可以解释一下

文中引入1\*1的卷积，主要时用于降低计算量，提高运行速度，另外某些网络1\*1的卷积可以用来降维

补充：(1)之前准确率一直在75%,可能是因为数据量的问题，用25000的猫狗数据进行分类，准确率轻松打到90%以上。

(2)代码中添加了用softmax进行分类的代码，详细看注释的代码，感觉二分类还是sigmod的精度高一些

[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3UwMTI2MDE2NDc_size_16_color_FFFFFF_t_70]: /images/20220215/3116c4ac63944bbeb54eed47decc2d8c.png